Segmentation of Hard exudates for the detection of Diabetic Retinopathy with RNN based sematic features using fundus images

Abstract

One of the diabetes complications that affects the eyes is Diabetic retinopathy and it is caused by blood vessels damage in the retina. The DR can be detected by finding the Hard Exudate present in it. The deep networks are becoming more deeper and more complex. So that adding a greater number of layers to a neural network can make it stronger for image related tasks. But the main drawback in adding more layers is that, it may greatly reduce the accuracy of the image and also the data models are complex. In order to overcome this drawback, Recurrent Neural Network can be introduced. The main aim for applying the recurrent neural network is that it can model a collection of records in such a way that each pattern is assumed to be dependent on the previous one. It can process inputs of any length. Even if the input size is large, the model size will not change. It makes the training process faster and attain more accuracy while compared to other neural networks. It greatly reduces the loss of accuracy because each lower knows the information of the top layers while updating the weights. This Recurrent has a greater number of parameters, so it is obvious that it can produce better result as compared to other net with the accuracy of 97.28%.